Research On CPU+GPU Heterogeneous Parallel Computing And Its Application In Compressible Flow

Computational Fluid Dynamics(CFD)is a subject that studies various flow phenomena and laws by numerically solving the flow control equation.CFD involves computer science and technology,fluid mechanics,computational mathematics and other professional research fields.In the field of aerospace,CFD has become one of the three research means to obtain the aerodynamic/thermal data of hypersonic vehicle and carry out the basic research of hypersonic flow.As the complexity of solving problems increases,geometric shapes,numerical methods,physical and chemical models,etc.become increasingly sophisticated and complex,CFD has a higher demand for computing resources.In recent years,GPU has developed rapidly,and the computing system based on CPU/GPU heterogeneous architecture has become the main way to construct large-scale supercomputer systems.The rapid development of computer performance brings new opportunities and challenges to CFD field.This paper focuses on the parallelization of compressible flow CFD algorithm in CPU/GPU heterogeneous computing system and compressible flow CFD applications.The research is mainly divided into two parts:First,performance research of commonly used flux splitting formats in the compressible field under GPU architecture.Taking one-dimensional shock tube as an example,a one-dimensional Euler solver is written using the CUDA programming model,and the Van Leer format,Roe format and AUSMPW+ format commonly used in the compressible field are analyzed on the CPU/GPU heterogeneous architecture.The results show that the Roe format has a good effect on flow field analysis,and has high acceleration performance on heterogeneous platforms,and has a good application prospect.Conditional branching in the scheme construction seriously affects the acceleration performance of the scheme.When constructing numerical schemes,reducing conditional branching reasonably will greatly improve the acceleration performance of the scheme on heterogeneous platforms.Second,based on the FR format of heterogeneous platform compressible stream solver.A parallel algorithm of GPU compressible flow was studied for flux reconstruction scheme FR,which is currently suitable for unstructured grids.Firstly,a parallel computing mode based on CPU/GPU heterogeneous computing system is proposed according to the characteristics of heterogeneous computer and CFD computing process.Implement parallel task partitioning based on FR format.Then,the parallel tasks are classified according to the types of computing points,and Kernel function is constructed.According to the characteristics of GPU data access,an array storage structure based on FR format is proposed.Finally,the effectiveness of the GPU algorithm was verified by an example of flow around a cylinder,and its acceleration ratio was evaluated.The proposed algorithm was successfully applied to the numerical simulation of a hypersonic lifting body.The experimental results show that the designed heterogeneous compressible flow solver can achieve a maximum acceleration ratio of more than 50 times on a single GPU compared with a single core CPU,and can be applied to complex hypersonic flow simulation,which has a certain engineering practical value.